What If Netscape Did SSL the HTTP Way?
6 points by cript0nauta
6 points by cript0nauta
Just want to point out that "connect in plaintext first, then upgrade" is a recipe for security vulnerabilities. If you want to know how that looks, just look at STARTTLS (I co-authored that paper): https://nostarttls.secvuln.info/
The HTTP-upgrade-to-TLS thing mentioned (RFC 2817) that is still implemented by some obscure HTTP servers is also flawed: https://opossum-attack.com/
The parallel universe gives that up. Its security is opportunistic: the browser asks whether the server can do encryption and uses it if so. That’s great, right up until someone hides the opportunity before the handshake completes. The negotiation happens in the clear, so an attacker sitting in the middle doesn’t need to break the encryption. They just stop it from ever starting. Anyone who has set up STARTTLS for email will recognize this one.
That's the obvious flaw, but you can sorta fix that by enforcing the STARTTLS upgrade and refuse to connect if it doesn't happen. But there's a whole can of worms of additional problems. You now have a complex state transition where your server and client have to know which part of the communication is before and after the transition.
Yeah, any interloper can just foul the necessary handshake requests and force a fallback to unencrypted HTTP. We had the same problem when people proposed adding facultative TLS to Gopher. The better solution was to just leave that to the Geminists.
And there was Opportunistic Encryption in HTTP too, for some brief time. See https://en.wikipedia.org/wiki/Opportunistic_encryption.
It’s a mistake to push TLS up the stack: it’s much better to move the security layer as low as possible (like QUIC) so that it can secure the transport, and so that different application protocols can share the security layer. TLS has been very good at application neutrality: basically only ssh has its own security layer and that’s largely due to it being too early.
Another interesting parallel universe is one in which IPSEC had been usable by applications without cooperation from the kernel, and supported single-end authentication with a PKI. (I don’t know enough about IPSEC’s key distribution and authentication model, but I guess that’s another reason it’s only used in VPNs.)
they can’t make one that’s valid for shop.example.com without a CA’s help, and if they can get one of those, you have bigger problems.
I used to share this stance before eMudhra.
What’s the alternative?
I just looked at the number of DNS resolutions my resolver has performed over the last week and it’s frankly staggering. And somebody has to provide a trust root for all of those - there’s no world I can think of where TOFU or WOT would scale to it.
WebPKI sucks, but it seems like the least sucky option.
The premise seemed interesting to me but the page is completely AI-generated unfortunately.